Adds Adafactor to the docs and slightly fixes the formatting

docs/source/main_classes/optimizer_schedules.rst

@@ -13,6 +13,11 @@ The ``.optimization`` module provides:
 .. autoclass:: transformers.AdamW
     :members:
 
+``AdaFactor`` (PyTorch)
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.Adafactor
+
 ``AdamWeightDecay`` (TensorFlow)
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 

src/transformers/optimization.py

@@ -328,31 +328,57 @@ class Adafactor(Optimizer):
     *warmup_init* options. To use a manual (external) learning rate schedule you should set `scale_parameter=False` and `relative_step=False`.
 
     Arguments:
-        params (iterable): iterable of parameters to optimize or dicts defining parameter groups
-        lr (float, optional): external learning rate (default: None)
-        eps (tuple[float, float]): regularization constants for square gradient
-            and parameter scale respectively (default: (1e-30, 1e-3))
-        clip_threshold (float, default 1.0): threshold of root mean square of final gradient update
-        decay_rate (float, default: -0.8): coefficient used to compute running averages of square
-        beta1 (float): coefficient used for computing running averages of gradient
-        weight_decay (float, default=0): weight decay (L2 penalty)
-        scale_parameter (bool, default: True): if True, learning rate is scaled by root mean square of
-        relative_step (bool, default: True): if True, time-dependent learning rate is computed instead of external learning rate
-        warmup_init (bool, default: False): time-dependent learning rate computation depends on whether warm-up initialization is being used
+        params (:obj:`Iterable[torch.nn.parameter.Parameter]`):
+            Iterable of parameters to optimize or dictionaries defining parameter groups.
+        lr (:obj:`float`, `optional`):
+            The external learning rate.
+        eps (:obj:`Tuple[float, float]`, `optional`, defaults to (1e-30, 1e-3)):
+            Regularization constants for square gradient and parameter scale respectively
+        clip_threshold (:obj:`float`, `optional`, defaults 1.0):
+            Threshold of root mean square of final gradient update
+        decay_rate (:obj:`float`, `optional`, defaults to -0.8):
+            Coefficient used to compute running averages of square
+        beta1 (:obj:`float`, `optional`):
+            Coefficient used for computing running averages of gradient
+        weight_decay (:obj:`float`, `optional`, defaults to 0):
+            Weight decay (L2 penalty)
+        scale_parameter (:obj:`bool`, `optional`, defaults to :obj:`True`):
+            If True, learning rate is scaled by root mean square
+        relative_step (:obj:`bool`, `optional`, defaults to :obj:`True`):
+            If True, time-dependent learning rate is computed instead of external learning rate
+        warmup_init (:obj:`bool`, `optional`, defaults to False):
+            Time-dependent learning rate computation depends on whether warm-up initialization is being used
 
     This implementation handles low-precision (FP16, bfloat) values, but we have not thoroughly tested.
 
     Recommended T5 finetuning settings:
-        scheduled LR warm-up to fixed LR, disable relative updates, use clip threshold: https://arxiv.org/abs/2004.14546
-        Adafactor(model.parameters(), lr=1e-3, relative_step=False, warmup_init=True)
-        Alternatively, relative_step with warmup_init can be used.
-        Training without LR warmup or clip threshold, is not recommended. Additional optimizer operations like gradient clipping, should not be used alongside Adafactor.
+        - Scheduled LR warm-up to fixed LR
+        - disable relative updates
+        - use clip threshold: https://arxiv.org/abs/2004.14546
+
+        Example::
+
+            Adafactor(model.parameters(), lr=1e-3, relative_step=False, warmup_init=True)
+
+        - Alternatively, relative_step with warmup_init can be used.
+        - Training without LR warmup or clip threshold is not recommended. Additional optimizer operations like
+          gradient clipping should not be used alongside Adafactor.
 
     Usage::
+
         # replace AdamW with Adafactor
-        optimizer = Adafactor(model.parameters(), lr=1e-3, eps=(1e-30, 1e-3), clip_threshold=1.0,
-            decay_rate=-0.8, beta1=None, weight_decay=0.0, relative_step=False,
-            scale_parameter=False, warmup_init=False,)
+        optimizer = Adafactor(
+            model.parameters(),
+            lr=1e-3,
+            eps=(1e-30, 1e-3),
+            clip_threshold=1.0,
+            decay_rate=-0.8,
+            beta1=None,
+            weight_decay=0.0,
+            relative_step=False,
+            scale_parameter=False,
+            warmup_init=False
+        )
     """
 
     def __init__(